[AMD] add slicing async-copy-local-to-global

Author: ravil-mobile

lib/Analysis/AxisInfo.cpp

@@ -3,6 +3,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 
+#include "third_party/amd/include/Dialect/TritonAMDGPU/IR/Dialect.h"
 #include "triton/Analysis/AxisInfo.h"
 #include "triton/Dialect/Triton/IR/Dialect.h"
 #include "triton/Dialect/Triton/IR/Utility.h"
@@ -1043,6 +1044,7 @@ AxisInfoAnalysis::AxisInfoAnalysis(DataFlowSolver &solver)
                   CastOpAxisInfoVisitor<arith::TruncIOp>,
                   CastOpAxisInfoVisitor<triton::gpu::ConvertLayoutOp>,
                   CastOpAxisInfoVisitor<mlir::UnrealizedConversionCastOp>,
+                  CastOpAxisInfoVisitor<triton::amdgpu::ExtractSliceOp>,
                   CastOpAxisInfoVisitor<triton::BitcastOp>>();
   visitors.append<MakeRangeOpAxisInfoVisitor>();
   visitors.append<PoisonOpAxisInfoVisitor>();

lib/Analysis/CMakeLists.txt

@@ -17,4 +17,5 @@ add_triton_library(TritonAnalysis
   TritonIR
   TritonGPUIR
   TritonNvidiaGPUIR
+  TritonAMDGPUIR
 )